Deflation subspace generation failure

Summary

While running at lighter quark masses we sometimes encounter an issue that the deflation subspace fail to generate. This is particularly prevalent close to T_c for some reason which originally had us thinking that it is a valid physical phenomenon. This is most likely not the case as one should always be able to generate the subspace, whether it is a good subspace or not.

It seems as if the issue can be somewhat ameliorated by making the deflation blocks larger, implying that there is something wrong with the boundaries of the deflation blocks. This is probably another anisotropy issue, however it is far from trivial.

Traceback

• Error code -1 from dfl_modes in modules/dfl/dfl_modes.c
• This status is set by dfl_smooth_fields
• Where it originates from ltl_gcr defined in modules/dfl/ltl_gcr.c
• Finally this comes from fgcr4vd defined in modules/linsolv/fgcr4vd.c, the error means that the CG algorithm failed to converge